Evaluating CPU cores for optimisation - page 2

 
Aleksey Vyazmikin:

8 passes 4 agents, kind of analogous to hypertrading - according to the number of threads.

Enable"Full Optimisation Log" to see the time of each pass.

Done:

2019.08.10 00:00:11.960 Core 4  pass 4 returned result 1001000.00 in 0:01:47.867
2019.08.10 00:00:12.838 Core 2  pass 2 returned result 1001000.00 in 0:01:48.960
2019.08.10 00:00:13.345 Core 1  pass 0 returned result 1001000.00 in 0:01:49.308
2019.08.10 00:00:13.914 Core 3  pass 6 returned result 1001000.00 in 0:01:49.925
2019.08.10 00:02:02.167 Core 4  pass 5 returned result 1001000.00 in 0:01:50.207
2019.08.10 00:02:03.038 Core 2  pass 3 returned result 1001000.00 in 0:01:50.198
2019.08.10 00:02:04.071 Core 1  pass 1 returned result 1001000.00 in 0:01:50.726
2019.08.10 00:02:05.018 Core 3  pass 7 returned result 1001000.00 in 0:01:51.102
2019.08.10 00:02:05.018 Tester  optimization finished, total passes 8
2019.08.10 00:02:05.028 Statistics      optimization done in 4 minutes 32 seconds
2019.08.10 00:02:05.028 Statistics      shortest pass 0:01:47.867, longest pass 0:01:51.102, average pass 0:01:49.786
This time it took about twice as long, which I think is logical (and Task Manager showed 57% CPU load).
 
Serhii Shevchuk:

Done:

This time it took about twice as long, which makes sense in my opinion (and the task manager showed 57% CPU load).

The results are very interesting, you can see that the agents are faster, but it turns out that this variant is slower in aggregate. The situation is different on full version of EA - and here I'm still guessing if it's OS, EA size (increase in number of sheets (comparison statements) slows down a lot, which is nonsense! As soon as I run out of optimizations on almost similar processor we will be able to compare results!

Added: Is the processor overclocked or not?
 
Aleksey Vyazmikin:
Is the processor overclocked or not?

No, the motherboard is primitive, it has overclocking problems. And what's more, the cooler is stock.

 
Serhii Shevchuk:

No, the motherboard is primitive, it has overclocking problems. And what's more, the cooler is stock.

I've got one of these - in principle it runs... But the mother is crappy in terms of power supply - I gave it to repair - then it sparks, then the cooler gets up (amazingly, but without cooler the CPU didn't shut down, but continued to optimize, though I trolled at 80 degrees and then raised the frequency again, but I have a tower there).

However, after the last repair the optimisation is faster (by 30%!), compared to the same processor, but on a different motherboard, which surprised me.
 
Aleksey Vyazmikin:

... or working with files.

At what level? I have an SSD system drive, but the metatrader (actually, the whole AppData folder) has been moved to HDD (liked to eat hundreds of gigabytes).

Don't know if it matters, but it's worth keeping in mind, just in case.

Added: I'll throw in the test results on some ultra-poor hardware now. Just for fun. I'm curious to see what happens.
 
Serhii Shevchuk:

At what level? I have SSD system disk, but metatrader (actually, the whole AppData folder) has been moved to HDD (liked to eat hundreds of gigabytes).

I don't know if it's important, but it's worth keeping in mind, just in case.

This version doesn't work with files, I mean the one I use in my work - it reads data from file, but here it just generates random numbers. Although, the file is not large and probably sits in RAM, which should not slow down the process, we'll see.

 
Serhii Shevchuk:
Added: I'll throw in the test results on some ultra-poor hardware. Just for fun. I'm curious to see what happens.

Let's

 
Aleksey Vyazmikin:

Let's

So. Iron:

celeron

Since the Celeron is basically hyperthreading-free, I ran two optimisation passes on the two cores available.

Terminal version:

2019.08.10 00:37:37.560 Terminal        MetaTrader 5 x64 build 2093 started (MetaQuotes Software Corp.)
2019.08.10 00:37:37.564 Terminal        Windows 10 (build 18362) x64, IE 11, RDP, UAC, Intel Celeron  847 @ 1.10 GHz, Memory: 5131 / 7871 Mb, Disk: 106 / 148 Gb, GMT+2

Results:

2019.08.10 01:04:42.229 Core 1  pass 0 returned result 1001000.00 in 0:09:42.962
2019.08.10 01:04:44.237 Core 2  pass 1 returned result 1001000.00 in 0:09:43.672
2019.08.10 01:04:44.237 Tester  optimization finished, total passes 2
2019.08.10 01:04:44.248 Statistics      optimization done in 9 minutes 45 seconds
2019.08.10 01:04:44.248 Statistics      shortest pass 0:09:42.962, longest pass 0:09:43.672
 
Serhii Shevchuk:

I'm curious to see how it turns out.

Please test this version on FX with 4 and 8 agents. The only difference in versions is in addition of if statements, which are nested in groups and should not degrade performance significantly in my opinion, but on laptop agents work 3 minutes faster on small version and there is an idea that small EA fits in CPU cache and due to this it works faster.

Files:
 
Serhii Shevchuk:

So. Hardware:

As the Celeron is in principle without hypertrading, I ran two optimisation passes on the two cores available.

Terminal version:

Results:

I thought I could get a faster CPU (buy it at AVito or Ali) and it would be OK.